Research Online

Logo

Goldsmiths - University of London

Time-dependent series variance learning with recurrent mixture density networks

Nikolaev, Nikolay; Peter, Tino and Evgueni, Smirnov. 2013. Time-dependent series variance learning with recurrent mixture density networks. Neurocomputing, 122, pp. 501-512. ISSN 0925-2312 [Article]

No full text available

Abstract or Description

This paper presents an improved nonlinear mixture density approach to modeling the time-dependent variance in time series. First, we elaborate a recurrent mixture density network for explicit modeling of the time conditional mixing coefficients, as well as the means and variances of its Gaussian mixture components. Second, we derive training equations with which all the network weights are inferred in the maximum likelihood framework. Crucially, we calculate temporal derivatives through time for dynamic estimation of the variance network parameters. Experimental results show that, when compared with a traditional linear heteroskedastic model, as well as with the nonlinear mixture density network trained with static derivatives, our dynamic recurrent network converges to more accurate results with better statistical characteristics and economic performance.

Item Type:

Article

Identification Number (DOI):

https://doi.org/10.1016/j.neucom.2013.05.014

Keywords:

Mixture density neural networks; GARCH models; Real-time recurrent learning algorithm

Departments, Centres and Research Units:

Computing

Dates:

DateEvent
25 December 2013Published

Item ID:

10591

Date Deposited:

02 Sep 2014 14:56

Last Modified:

20 Jun 2017 11:17

Peer Reviewed:

Yes, this version has been peer-reviewed.

URI:

http://research.gold.ac.uk/id/eprint/10591

Edit Record Edit Record (login required)